MARKOV RANDOM FIELDS PRE-WARPING TO PREVENT COLLUSION IN IMAGE TRANSACTION WATERMARKING

摘要

Transaction watermarks can be used to track back the source of unauthorized content. They individualize images by embedding unique watermark identifiers in each copy. However, collusion attack is a fundamental security problem for transaction watermarking. Since each copy is differently watermarked for different recipients, one or many adversaries may collude together by using multiple copies to compose a new copy. Such copy may contain no valid identifier to avoid being traced, or even a new identifier that frames an innocent user. In this paper, we propose a novel collusion-resilience pre-warping mechanism for image watermarking by using a Markov random displacement field. The typical image quality impairment caused by pre-warping is alleviated by applying Markov Random Field (MRF) and automated image post-processing. Besides, the proposed approach is independent of the watermarking algorithm used and the watermark signal. It can therefore be considered as an additional security layer to improve the collusion resilience for an existing watermarking system. Experimental results demonstrate the effectiveness of the proposed pre-warping solution against various collusion attacks.